Skripsi
ANALISA KEPADATAN LALU LINTAS DI KOTA PALEMBANG MENGGUNAKAN METODE CNN-GRU BERDASARKAN DATA MEDIA SOSIAL
Traffic congestion in Palembang City significantly impacts residents' mobility and urban productivity. This study develops a traffic density classification system based on TikTok social media text using a hybrid Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) method. A total of 6,446 captions and comments were collected via Apify and processed into 2,260 relevant records, labeled into three classes: negative (congested), neutral (moderately busy), and positive (clear). Class imbalance was addressed using random deletion, random swap, and synonym replacement. The CNN-GRU model was trained with categorical focal loss and the Adam optimizer, achieving an accuracy of 89.82% and a weighted F1-score of 0.8957. The negative class attained the highest F1-score of 0.94. These results confirm that CNN-GRU is effective for classifying traffic conditions from social media text, offering a viable AI-based traffic monitoring alternative.